The Human Capital Study 2002–04: Tracking, data collection, coverage, and attrition

Abstract

Between 2002 and 2004, the Institute of Nutrition of Central America and Panama (INCAP), in collaboration with Emory University, the International Food Policy Research Institute (IFPRI), and the University of Pennsylvania, re-surveyed young Guatemalan adults who had, as children, been participants in a nutrition supplementation trial conducted by INCAP between 1969 and 1977. This “Human Capital Study 2002–04” complements and extends data obtained in previous studies by collecting new information on measures of physical health and well-being, schooling and cognitive ability, wealth, consumption and economic productivity, and marriage and fertility histories. This paper describes the study domains and data collection procedures. Among 2,393 members of the original sample, 1,856 (77%) were targets for enrollment. Response rates varied by gender, current place of residence, and domain of data collection, with 80% of males and 89% of females completing at least one data collection instrument. Attrition was not random and appears to be associated with a number of initial characteristics of individuals and their households that should be controlled for in future analyses. We conclude that data collection was successful and data quality is high, facilitating the successful undertaking of our planned investigation of important study hypotheses.

Introduction

Between 2002 and 2004, the Institute of Nutrition of Central America and Panama (INCAP), in collaboration with Emory University, the International Food Policy Research Institute (IFPRI), and the University of Pennsylvania, re-surveyed young Guatemalan adults who had, as children, been participants in a nutrition supplementation trial conducted by INCAP between 1969 and 1977 [1]. The present study, the Human Capital Study 2002–04, complements and extends data obtained in previous studies [1] by collecting new information on measures of physical health and well-being, schooling and cognitive ability, wealth, consumption and economic productivity, and marriage and fertility histories. The rationale and key objectives of the Human Capital Study are described in a companion paper [2]. Here we describe the procedures for tracing and contacting original sample members of the INCAP Longitudinal Study (1969–77) and for data management and the overall success in re-interviewing original study participants for the Human Capital Study from 2002–04.

Methods

Our approach to collecting data for the Human Capital Study required careful planning and execution. The tracing of former participants, the collecting and entering of data, the survey instruments that were implemented, and the mechanisms that were put in place to ensure data quality, are described below.

Tracing former participants

All 2,393 individuals born between 1962 and 1977 who participated in the INCAP Longitudinal Study were eligible for follow-up in the Human Capital Study. In 2003, during the mid-point of fieldwork, these individuals would have been between 26 and 41 years old.

Maps of the original study villages were obtained from the National Statistics Bureau and updated as needed. A census of all households in the original villages was implemented between January and April 2002. Sociodemographic information was collected from the entire population and records regarding mortality and migratory status of the 2,393 original sample members were updated. A list of “missing” sample members was created, and was then reviewed and corrected by questioning sample members’ relatives, peers, (former) neighbors, and five community leaders in each of the original villages. Original sample members who had moved away from their natal villages were classified as “migrants.” Research staff interviewed migrants’ family members and acquaintances to obtain information about the migrants’ current addresses and phone numbers, employers, work addresses, or general whereabouts. Flyers soliciting this information and invitation letters were also left with the relatives of migrants. The 102 (4%) sample members for whom no information was available even after these extensive efforts were classified as “untraceable.”

Sample members still living in the original villages were visited by a team of interviewers and invited to participate in the study. Those who accepted were asked to sign a written informed consent approved by the Emory University and INCAP Institutional Review Boards. The informed consent was updated before each session of data collection (described below).

While collecting data in the original study villages, we also attempted to contact and interview migrants who were visiting their natal villages—for example, during village feast days and other major holidays. Between January 2003 and April 2004, a two-person team (one man and one woman) traveled throughout Guatemala to attempt to locate other migrants. Migrants to nearby villages, Guatemala City, or other cities or towns in Guatemala were visited wherever they were living, and invited to participate. In addition, we used a “snowball’ approach whereby a list of still-missing original sample members was reviewed with each migrant located.

Data collection

Data collection was carried out between May 2002 and April 2004 (table 1). During the first year, data collection focused on sample members residing in the original villages, while from January 2003 to April 2004 the focus shifted to those residing elsewhere in El Progreso (the department in which the original study villages are located), in Guatemala City, and elsewhere in Guatemala.

TABLE 1.

Schedule of data collection for the Human Capital Study 2002–04

2002			2003			2004
Jan–April	May–August	Sept–Dec	Jan–April	May–August	Sept–Dec	Jan–April
Preparatory phase	Sample members residing in original villages			Sample members residing elsewhere in Guatemala
	Module 1	Module 2	Module 3	All modules
Socio-anthropologic study Socio-demographic characteristics (Census)	Literacy Schooling Reading skills Intelligence Individual diet Physical activity	Anthropometry Marriage and assets history Household expenses	Reproductive history Medical history/physical exam Blood test	Socio-demographic characteristics (Census) Literacy Schooling Reading skills Intelligence Individual diet Physical activity Anthropometry Marriage and assets history Household expenses Reproductive history Medical history and physical exam Blood test Economic activity
		Module 4
		Economic activity

Data were collected from residents of the original study villages between May 2002 and April 2003, in four (partly overlapping) modules, each fielded for about 4 months each (tables 1 and 2). Each module typically required two to three interview visits to complete. To limit respondent fatigue, interviews were scheduled at least 4 weeks apart. Most interviews were conducted in respondents’ homes with the exception of the physical tests that were done at local INCAP headquarters.

TABLE 2.

Study domain and data collection of the Human Capital Study 2002–04

Study domain	Data collection
Preparatory phase
Socio-anthropologic Socio-demographic characteristics (Census 2002)	Current and past education and health facilities, physical infrastructure, public services, and programs, events that might have affected human capital or economic productivity Demographics: Date of birth, gender, migration status, religion, literacy, schooling, occupation, and mortality. House structure: Type of walls, roof, and floor, number of rooms, and availability of latrines was obtained by interviewers’ observation of the house. Family possessions: Household appliances, animals, land tenure.
Module 1
Literacy Schooling Reading performance Intelligence Individual diet Physical activity	Literacy, knowledge of letters, syllables, words and general reading skills, including phrases and short sentences. Schooling histories and scholastic achievement. Reading comprehension and vocabulary section of the Interamerican Reading Series Raven’s Standard Progressive Matrices test. A semi-quantitative food frequency questionnaire Timing, duration, nature and intensity of key events (getting ready for work, household chores, means of transportation to work, type of activity at work, duration of working day and types of recreational activities).
Module 2
Anthropometry Marriage and assets history Household expenses	Weight, height, seated height, knee height, triceps and subscapular skin folds, waist, and hip circumferences. Marital status and history; characteristics and family background of spouses; and assets brought by both parties to the household and their source. Food and non-food household expenditures including education and health expenditures.
Module 3
Reproductive history Medical history and physical exam Blood tests	Males: marriage history and knowledge and practices relating to contraceptives. Females: age at menarche, date of the last menstrual period, number of pregnancies and parity, obstetrical history of each pregnancy. Personal and family history of health, smoking, drinking, medication and drug consumption and any current symptoms of disease and the physical examination including body temperature, heart and respiratory rates, blood pressure and examination of eyes, ears, neck, chest, thorax (heart rates and murmurs, respiratory tremors), abdomen (gastrointestinal murmurs) and limbs (deformities and limitations) examination, physical fitness (step test, hand strength, flexibility). Total cholesterol, HDL-cholesterol, triglyceride, glucose, hemoglobin
Module 4
Current economic activity Other	Employment status, jobs held and migratory status, job description, contractual arrangements, incomes, wages if paid, employer characteristics, land operated, crops grown, value of harvest, income generated from non-agricultural own business activities, transfers Self-esteem, social capital, and remittances

Data were collected from migrants between May 2003 and April 2004. By April 2003, each interviewer had mastered at least four different areas of data collection, enabling two interviewers and the physician to collect all the data related to a single respondent in a single session. Sample members who lived in Guatemala City were invited to INCAP headquarters for interviews and examinations. Transportation was arranged if needed. Migrants living in Guatemala City who did not come to INCAP were visited by a team of interviewers and interviewed and examined in their homes.

Data management

Double-data entry (using Epi-Info, version 6.04) was carried out and, in contrast to previous studies of this sample, was done in the study field headquarters. Data were typically entered within a few days of the interview, though occasionally data entry was delayed up to 4 weeks after data collection. The “Validate” option within Epi-Info was used to compare files. The resulting files therefore accurately reflect the information registered in the forms. Cleaning routines were developed and applied. Basic checks such as range values and logical consistency across variables were conducted every few days and queries were resolved by the supervisors in the villages. When possible, potential errors were sent to the field for review, with the supervisor authorized to correct coding errors.

A data back-up system was implemented at study field headquarters, as well as at INCAP’s headquarters in Guatemala City. Data files were sent to Emory University a few weeks after data collection was finished in the original villages and after migrants’ data collection was completed. Emory University acted as the central repository for the data and syntax files. Conventions were established to standardize file names and to assign variable labels across all data files that were created. A second set of more complex data cleaning routines were developed and performed by researchers with expertise in specific areas; potential errors and inconsistencies were discussed with the field director and field supervisor and necessary corrections incorporated in the data set.

Data collected during the follow-up

In early 2002, a socio-anthropological study [3, 4] was implemented to describe current and past educational and health facilities, physical infrastructure, public services, and programs that operated in the study area, as well as important events that might have affected human capital or economic productivity; this information helped us to design different forms and questionnaires. Individual- and household-level data were collected related to sociodemographic information, literacy, schooling, reading skills, intelligence [5], individual diet, physical activity [6], anthropometry, medical history and physical exam, blood tests [7], reproductive history [8], marriage and assets histories [9]; household expenses [10]; economic activity [11] and other information. A more detailed description of study domains (table 2) and data collection methods is provided in the respective articles in this special issue [7–11]. To ensure comparability across previous studies, all questionnaires and study domains were reviewed and followed where possible; instruments for both old and new domains of interest were reviewed, updated, and validated by first applying them to citizens from nearby similar villages.

Interviewer training and field quality control

Code books were prepared for each form, questionnaire, or test. Each team of interviewers was trained and interviewing techniques and interpretations were standardized in one or two study domains in the two weeks before the module implementation. At least two re-standardization exercises were done within the module implementation; the interviewers performed a supervised reading of the code books and cross interviews applying to citizens from nearby similar villages; the interviewers’ field experiences were shared with the whole field work team and lessons learned were summarized. All forms were reviewed by the interviewers after the completion of the interviews. The field workers were directly supervised, with 10% of the interviews of each study domain observed by the field supervisor. The quality of data collection was extensively monitored during data collection in the original villages; repeated measurements or cross interviews were done by the supervisors or other interviewers in at least 5% of the interviews and the percentage of agreement among interviewers or supervisors usually exceeded 95%. The results of the supervisors’ observations and the repeated measurements or cross interviews were used to field-train interviewers and improve ongoing data collection. The field supervisor reviewed 40% of the forms filled in the original villages and 60% of the forms filled elsewhere in Guatemala, looking for non-permissible data, missing information, and inconsistencies between questions. Forms with inconsistencies or missing data were re-sent to the field to be corrected. After they were reviewed, the forms were delivered to the data center established at INCAP headquarters in each study village.

Results

Tracing and enrollment in the targeted sample

Of the original sample of 2,393 individuals, 272 (11%) had died by 2002, the majority due to infectious diseases in early childhood (fig. 1). For 102 (4%) sample members, even after extensive investigation, no information could be found. These 102 “untraceable” sample members may or may not be alive and, if alive, may or may not be living in Guatemala; 71 of these 102 also did not appear in the 1975 census. It is probable that they (and their families) had moved away nearly 30 years ago, and we have been unable to trace them since. Of the remaining 2,019 sample members, 163 (8%) had migrated out of the country, three-quarters to the United States. If all 102 untraceable sample members were either dead or living outside of Guatemala, the true targeted population would comprise 1,856 original sample members (77% of the original cohort); if all 102 were alive and living in Guatemala, there would be 1,958 (82%) potential sample members.

FIG 1.

Tracking of the Human Capital Study 2002–04 sample. Of the original sample of 2,393 individuals in the 1969–77 INCAP Longitudinal Study, 272 (11%) had died by 2002, and 102 (4%) sample members were not found. Of the remaining 2,019 sample members, 163 (8%) had migrated out of the country. The true targeted population for the Human Capital Study 2002–04 comprises 1,856 (77%) original sample members.

Coverage and attrition

In what follows, we base coverage calculations on the smaller target sample size of 1,856. Obviously, coverage rates would be lower than those reported below if we were to treat untraceable sample members as a part of the target sample, and even lower if we were to treat all original 2,393 original sample members as the target sample. Doing so, however, would lead to an unfair assessment of the success of fieldwork activities in isolation, since only the 1,856 individuals were targeted. We return to a consideration of the full original sample (2,393) in the examination of attrition below.

For the 1,856 traceable sample members living in Guatemala, 1,054 (58%) finished the complete battery of applicable interviews and measurements and 1,570 (85%) completed at least one interview (table 3). For two-thirds of the 286 (15%) who completed no interviews, while we learned they were alive and living in Guatemala, we were unable to obtain a current address and therefore could not make contact. As a result, the effective refusal rates for any participation among those whom we were able to contact were low, around 5%.

TABLE 3.

Coverage in the Human Capital Study 2002– 04 by type of data collection: gender and current residence

Current residence	Men (n = 937)						Women (n = 919)
	Original or nearby villages (n = 651)		Guatemala City (n = 197)		Elsewhere in Guatemala (n = 89)		Original or nearby villages (n = 616)		Guatemala City (n = 222)		Elsewhere in Guatemala (n = 81)
	N	%a	N	%a	N	%a	N	%a	N	%a	N	%a
Literacy/schooling	540	83	100	51	43	48	583	95	141	64	62	77
Reading performanceb	433	78	98	50	33	42	454	92	133	62	45	69
Intelligence	528	81	100	51	43	48	578	94	140	63	61	75
Individual diet	552	85	101	51	44	49	583	95	145	65	61	75
Physical activity	552	85	101	51	44	49	584	95	145	65	61	75
Anthropometry	482	74	101	51	44	49	556	90	146	66	60	74
Married and assets history	490	75	102	51	44	49	562	91	147	66	61	75
Household expenses	594	91	107	54	50	56	571	93	143	64	59	73
Reproductive history	502	77	101	51	44	49	569	93	149	67	61	75
Medical history and physical exam	448	69	100	51	44	49	550	90	146	66	61	75
Blood test	353	54	95	48	44	49	503	82	146	66	60	74
Current economic activity	510	78	99	50	44	49	569	92	138	62	62	77
Total coverage
Completed all forms	280	43	80	41	42	47	473	77	123	55	56	69
Completed at least one form	597	92	110	56	45	51	598	97	157	71	63	78
No address	0	0	73	84	39	89	2	11	56	86	17	94
Completed all interviews	394	61	87	44	43	48	531	86	130	59	58	72
Completed all physical exams	308	47	89	45	43	48	486	79	133	60	57	70

^aCoverage = collected (measured) / eligible.

^bCoverage is calculated only for those eligible subjects.

Coverage varied by gender, data domain, and current place of residence (table 3). Coverage was lower for men than for women—43% of men and 71% of women completed all the forms, and 80% versus 89%, respectively, completed at least one form (data not shown). Completion rates were higher for instruments involving only interviews as compared with those involving physical measurements; 56% of males and 78% of females responded to all interviews, while 47% of males and 73% of females completed all physical measurements (data not shown). No differences in response rates were found across birth year cohorts or tertiles of 1,975 parental SES scores, for men or women (data not shown).

Of the 1,856 target sample members, 1,113 (60%) currently reside in their native village, 154 (8%) live in nearby villages (almost all of them within 10 km from the original villages), 419 (23%) live in Guatemala City, and the remaining 170 (9%) live in other cities or towns elsewhere in Guatemala (fig. 1).

Among residents of the original (or nearby) villages, 77% of the females and 43% males responded to all the forms and 97% and 92%, respectively, responded to at least one form. In Guatemala City, 71% of females and 56% of males responded to at least one form. Overall, then, higher response rates were achieved for sample members living in the original villages. Due to the logistics of data collection, however, which entailed multiple visits, there was higher variability in the response rates across instruments among sample members living in the original villages than those living in Guatemala City or elsewhere in Guatemala, where all data collection usually was conducted in a single visit. For both males and females, the blood test had the lowest response rate. For all data domains, response rates were higher for women than for men. No difference in response rates were found among data domains between those living in Guatemala City and living elsewhere in Guatemala.

The discussion of coverage demonstrates how successful the study was in interviewing those in the target sample, the result of the various methodologies for tracing that were put in place. Even if 100% of the target sample had been interviewed, however, the study would still have had substantial attrition, i.e., original sample members who were not interviewed in 2002–04, because not all original sample members were targeted. When making inference using these data, what is most important to the analyst is not coverage of the target sample, as defined above, but rather overall attrition. What are the characteristics of those individuals not interviewed, for whatever reason, in 2002–04? We use multinomial logistic regression to describe the factors associated with attrition, with the outcome as the dependent variable coded as at least one completed interview (the base category), dead, untraceable, migrated outside the country, and not interviewed (table 4). The latter group comprises those 286 in the operational target sample of 1,856 who were not interviewed in 2002–04.

TABLE 4.

Attrition analysis for the Human Capital Study 2002–04: Logistic regression estimates^a

	Dead			Untraced			Living outside Guatemala			Found, not interviewed
	Odds ratio	95% CI		Odds ratio	95% CI		Odds ratio	95% CI		Odds ratio	95% CI
Male	1.47b	1.12	1.93	0.65	0.39	1.07	1.25	0.90	1.75	1.85b	1.40	2.43
Year of birth	1.09b	1.06	1.13	1.02	0.95	1.09	0.98	0.94	1.02	1.02	0.99	1.05
Born in San Juan	1.09	0.74	1.62	1.10	0.58	2.07	0.61	0.34	1.07	0.83	0.57	1.22
Born in Conacaste	1.39	0.97	1.98	0.30b	0.14	0.65	1.64b	1.08	2.48	0.99	0.70	1.38
Born in Espíritu Santo	1.07	0.70	1.63	0.58	0.29	1.17	1.04	0.62	1.73	0.61b	0.40	0.93
Age of mother when born	1.00	0.97	1.02	1.02	0.97	1.06	0.99	0.96	1.02	1.01	0.98	1.03
Age of father when born	1.02b	1.00	1.05	0.99	0.95	1.03	0.99	0.96	1.02	1.00	0.97	1.02
Mother’s schooling	0.97	0.88	1.07	0.79	0.60	1.04	1.03	0.92	1.15	1.05	0.96	1.14
Father’s schooling	0.92b	0.84	0.99	1.28b	1.11	1.47	0.98	0.89	1.07	1.04	0.97	1.11
SES in 1975	0.99	0.90	1.09	0.72	0.51	1.02	1.14b	1.04	1.25	1.05	0.96	1.14
Missing SES	0.39b	0.20	0.76	1.50	0.54	4.13	0.56	0.27	1.18	0.44b	0.23	0.83

^aDependent variable is final HCS study status (reference category is ‘interviewed’). Reference categories for village of origin is Born in Santo Domingo. Regression also includes dummy control variables for indicator of whether mother’s age, father’s age, mother’s schooling, or father’s schooling are missing.

^bSignificant at p < .05.

Attrition in the sample is associated with a number of initial conditions, with the effects differing by the reason for attrition. Compared with those interviewed, men were more likely to have died or were not interviewed, and were slightly more likely to have migrated out of Guatemala (though this effect is not significant, p = .18). The association of later year of birth, i.e., younger age, with risk of death is counterintuitive and results from the inclusion in the study population of all children less than seven years in 1969. These represent the survivors of their respective birth cohorts, and hence they experienced a lower mortality rate (most of which is driven by infant mortality) compared with the later birth cohorts in the study who were followed from birth.

There was no significant pattern for death by village of birth, but those who were born in Conacaste were significantly less likely to be untraceable than those born in Santo Domingo and significantly more likely to be living outside Guatemala. Those born in Espíritu Santo and in the target sample were more likely to be interviewed compared with those from Santo Domingo; this is consistent with somewhat higher cooperation with the study team in Espíritu Santo, the least urbanized of the villages.

We next considered the role of a number of measures of family background on attrition in the sample, including parental age (at child birth), parental education, and a 1975 parental socioeconomic status (SES) score. Age of parents (at child birth) and maternal education have little effect on attrition. Fathers’ schooling, however, is associated with lower odds that the sample member was lost to death and higher odds that he or she was untraceable. To the extent that paternal education is an indicator of long-run resources in the household, these findings are consistent with the 2002–04 sample members being from better off households in which there were fewer deaths. Another indicator of resources is the 1975 parental household SES score, which increases significantly the odds of living outside of Guatemala. Having missing household SES also substantially decreases the odds of death and the odds of not being interviewed, conditional on being in the target sample.

To probe deeper into the extent to which the factors in table 4 or other factors vary across the different categories of attrition groups, we next examine the means and SDs for a variety of variables (table 5). While all but two of the variables show significant differences across categories using an analysis of variance (ANOVA) test (second to last column), fewer than half show significant differences when we compare means between the interviewed and all non-interviewed (whatever the reason) individuals together, using two sample t-tests or proportion tests as appropriate (final column). While there are significant differences across categories as indicated by the ANOVA tests, in many cases they are in opposite directions across categories within the not-interviewed group such that on average the group not interviewed is not significantly different from the group that was interviewed. This suggests that, while clearly not random, the average effects of attrition do not show obvious patterns of bias. Further, for those that do differ by either test, several do not appear to be very large differences. For example, the proportion living in Espíritu Santo varies by only seven percentage points across categories and the year of birth by only 2.0 years. However, household SES score in 1975 differs by a full SD across categories and the height-for-age z score by 0.6 SD, with those best-off living outside Guatemala or traced but not interviewed.

TABLE 5.

Mean and SDs of selected variables: by attrition category of the Human Capital Study 2002–04

	Interviewed	Not interviewed				ANOVA p	t-testap
		Dead	Untraced	Living outside Guatemala	Traced, not interviewed
Number of observations (1) if Santo Domingo (1) if San Juan (1) if Conacaste (1) if Espíritu Santo Year of birth	1570 0.26 0.23 0.30 0.21 1970.1 (4.2)	272 0.24 0.21 0.36 0.18 1971.5 (3.9)	102 0.35 0.31 0.12 0.22 1970.1 (3.5)	163 0.27 0.12 0.43 0.18 1969.5 (4.2)	286 0.34 0.20 0.32 0.15 1970.2 (4.3)	.02 < .01 < .01 .16 < .01	.11 .18 .19 .05 .05
(1) if male (1) if atole Age of mother when born (years)	0.48 0.53 27.6 (7.2)	0.59 0.58 28.3 (8.0)	0.43 0.43 25.8 (7.6)	0.55 0.55 26.4 (6.9)	0.65 0.52 27.2 (7.4)	< .01 .15 < .01	< .01 .92 .22
Age of father when born (years)	32.9 (8.7)	35.0 (9.5)	31.9 (9.0)	31.6 (8.1)	32.5 (8.7)	< .01	.51
Mother’s schooling (years completed)	1.3 (1.7)	1.1 (1.4)	0.9 (1.7)	1.5 (1.4)	1.5 (1.7)	.05	.99
Father’s schooling (years completed)	1.7 (2.2)	1.3 (1.7)	3.0 (3.4)	1.7 (2.2)	2.0 (2.4)	< .01	.90
Household SES score in 1975	−0.19 (1.67)	−0.27 (1.69)	−0.60 (1.40)	0.36 (1.84)	0.16 (1.86)	< .01	.02
(1) if missing 1975 SES score Height-for-age (HAZ) z score at 2 years	0.08 −2.47 (1.09)	0.13 −2.78 (0.98)	0.72 −2.37 (0.86)	0.16 −2.15 (0.88)	0.20 −2.18 (1.00)	< .01 < .01	< .01 .07
(1) if missing HAZ score	0.54	0.84	0.90	0.66	0.69	< .01	< .01

^aTwo sample t-test except where variable is a proportion, where it is a test of equality of proportions.

This analysis, while descriptive, underscores that attrition in the sample is not random. Analysts will need to consider its potential effects when using these data and, at a minimum, control for some of these initial factors.

Spouses of original sample members were also targeted for interview. Because the original study covered all children born in the villages over a 15-year period, many target sample members intermarried, with the consequence that in many instances, interviewing spouses meant interviewing an original sample member; there were 216 couples formed by two original study sample members. In the end, we interviewed 703 spouses who were not part of the original study, out of 1,038 projected. To project the total number of spouses who were not original sample members (in order to calculate coverage for them), we do the following within each location (original villages, Guatemala City, elsewhere in Guatemala). First we calculate the number of spouses identified in the field work who were not original sample members. Then, we divide by the number of original sample members interviewed. This fraction is then applied to the total number of original sample members (regardless of whether they were interviewed) to obtain the projected number of spouses who are not original sample members. Coverage of spouses was 84%, 45% and 53% for spouses living in original villages, Guatemala City and elsewhere in Guatemala, respectively—broadly similar rates to original male sample members.

Conclusion

The strategies implemented to track and locate former participants of the INCAP Longitudinal Study (1969–77) allowed us to trace and re-interview 84% of those individuals alive and known to be living in Guatemala, which was 66% of the original sample of 2,393. Individuals were found using information obtained in village censuses that focused on mortality and migratory status. Lists of missing sample members were produced and updated frequently and reviewed by relatives, peers, neighbors, and community leaders and other migrants. Flyers and invitation letters were left with migrants’ relatives. A system was put in place to contact (and at times interview) migrants during visits to their home village. Once we had a phone number or address list a two-person team telephoned or visited the place of work or the potential home address until contacting them. These strategies required substantial dedication and financial support but allowed a high level of coverage.

The Human Capital Study (2002–04) complements and extends the data collected in previous studies with information on physical health and well-being, schooling and cognitive ability, wealth, consumption and economic productivity, and marriage and fertility histories. The data collection was organized in such a way that training, standardization, and supervision ensured data quality and comparability across previous studies. As an innovation a computer center was established close to the original villages, which allowed data entry and cleaning a few weeks after the interviews and physical measurements were performed and conducting descriptive analysis within 6 months of the completion of data collection.

The coverage varied among data areas, gender, and current place of residence. However, more than 80% of males and 89% of females in the original sample who are known to have been alive and in Guatemala completed at least one form. We conclude that the strategies implemented for tracking sample members and those to perform data collection were fairly successful, which affects the coverage and attrition levels and allow us better to test our hypotheses in future studies.